Models · Announcements
Introducing Comet
2 August 2026
Comet is our speed model. It runs on every commit, catching slop, behavioural regressions, and the obvious security holes before they ever reach a branch.
The cheapest time to fix generated slop is the moment it is written. A placeholder section that survives one commit tends to survive the next twenty, and by the time anyone reads the diff properly it has load-bearing code built on top of it.
Comet exists to close that window. At 120 ms and 320 tokens per decision it is the fastest and cheapest model in the comparison below — fast enough to sit in a pre-commit hook or run inline in the editor — while still landing at 96.8% accuracy, ahead of every external model we test.
| Model | Token Usage | Time Taken | Accuracy |
|---|---|---|---|
 Nebula | 1,350 tokens | 780 ms | 93.6% |
| Comet | 320 tokens | 120 ms | 96.8% |
 Fable 5 | 1,420 tokens | 820 ms | 92.8% |
| Opus 5 | 3,600 tokens | 1,650 ms | 89.5% |
 GPT 5.6 Sol | 2,100 tokens | 1,240 ms | 84.2% |
 Gemini CLI | 2,850 tokens | 2,100 ms | 78.6% |
| Gemini 3.6 Flash | 890 tokens | 450 ms | 72.1% |
Performance and cost
Comet is not a smaller version of Nebula running the same passes badly. It scores the surface that actually changed, against the last known-good state of the flows that surface belongs to, which is why it can return inside a normal editor round trip.
The cost of that scope is cross-file reasoning. Comet will not find a vulnerability that only exists because two routes trust each other, and it does not guess at them — anything that needs a full trace is escalated to Nebula rather than answered cheaply.
Time Taken
Working with Comet
A Comet pass runs the same three stages as Nebula, scoped to the changed surface rather than the whole project.
- De-slop. Catches generated filler at the moment it is written — duplicate sections, placeholder copy, components that exist only because a template had them. Cheap enough to run on every save.
- Behaviour. Flags regressions against the last known-good state of a flow: a primary action that moved, a label that stopped matching its destination, a step that quietly got longer.
- Security. Blocks the high-frequency mistakes before they reach a branch — committed secrets, unguarded routes, unvalidated input — and escalates anything needing a deeper trace.
Review and safety
Comet only ever proposes. Nothing it produces reaches your codebase until you approve it, and every decision arrives with the reasoning that produced it.
Where Comet is not confident, it says so and escalates rather than answering anyway. A fast model that guesses is worse than no model at all.
Token Usage
Getting started
Comet is available in the CLI as a pre-commit hook or watch process, inline in Cursor and VS Code, and as the default check on every push through GitHub, Vercel, and Netlify. It is billed at $0.01 per commit and $0.15 for a full project sweep.
For release branches, security review, or the first pass over a project Behavr has never seen, send it to Nebula instead.